US10953891B2 - Method and system for providing an optimized control of a complex dynamical system - Google Patents

Method and system for providing an optimized control of a complex dynamical system Download PDF

Info

Publication number
US10953891B2
US10953891B2 US15/963,240 US201815963240A US10953891B2 US 10953891 B2 US10953891 B2 US 10953891B2 US 201815963240 A US201815963240 A US 201815963240A US 10953891 B2 US10953891 B2 US 10953891B2
Authority
US
United States
Prior art keywords
control
vector
dynamical system
heuristics
machine learned
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related, expires
Application number
US15/963,240
Other versions
US20190031204A1 (en
Inventor
Dirk Hartmann
Birgit Obst
Erik Olof Johannes Wannerberg
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Siemens AG
Original Assignee
Siemens AG
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Siemens AG filed Critical Siemens AG
Assigned to SIEMENS AKTIENGESELLSCHAFT reassignment SIEMENS AKTIENGESELLSCHAFT ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: Wannerberg, Erik Olof Johannes, OBST, Birgit, HARTMANN, DIRK
Publication of US20190031204A1 publication Critical patent/US20190031204A1/en
Application granted granted Critical
Publication of US10953891B2 publication Critical patent/US10953891B2/en
Expired - Fee Related legal-status Critical Current
Adjusted expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W50/06—Improving the dynamic response of the control system, e.g. improving the speed of regulation or avoiding hunting or overshoot
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/048—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators using a predictor
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
    • G05D1/02—Control of position or course in two dimensions
    • G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
    • G05D1/0212—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory
    • G05D1/0221—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory involving a learning process
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/33—Director till display
    • G05B2219/33037—Learn parameters of network offline, not while controlling system
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/33—Director till display
    • G05B2219/33041—Structure optimization and learning of artificial neural network by genetic algorithm
    • G05D2201/0213—

Definitions

  • the following relates to a system and method for providing optimized control of a complex dynamical system such as a vehicle using machine learned, scenario based control heuristics.
  • An aspect relates to providing a method system for controlling a complex dynamical system which is highly efficient and requires less computational resources for performing the control of the complex dynamical system.
  • Embodiments of the present invention provide according to a first aspect a method for performing an optimized control of a complex dynamical system using machine learned, scenario based control heuristics.
  • the method comprising the steps of: providing a simulation model for predicting a system state vector of said dynamical system in time based on a current scenario parameter vector and a control vector, using a Model Predictive Control, MPC, algorithm to provide the control vector at every time during a simulation of the dynamical system using said simulation model for different scenario parameter vectors and initial system state vectors, calculating for every simulated combination of scenario parameter vector and initial system state vector a resulting optimal control value by the Model Predictive Control, MPC, algorithm and saving the resulting optimal control value, generating machine learned control heuristics approximating the relationship between the corresponding scenario parameter vector and the initial system state vector for the saved resulting optimal control value using a machine learning algorithm, and using the generated machine learned control heuristics to control the complex dynamical system modelled by said simulation model.
  • MPC
  • the machine learning algorithm uses diffusion maps.
  • the machine learning algorithm uses diffusion maps with closed observables for approximating the dynamical system.
  • the machine learning algorithm uses support vector machines.
  • the generated machine learned control heuristics are transferred to a controller which controls online the dynamical system according to the transferred machine learned control heuristics.
  • the machine learned control heuristics comprise approximate rules for controlling the complex dynamical system modelled by said simulation model.
  • Embodiments of the present invention provide according to a further aspect a control heuristic generation platform.
  • Embodiments of the present invention provide according to the second aspect a control heuristic generation platform for providing machine learned control heuristics used for controlling a dynamical system modelled by a simulation model f stored in a model storage and adapted to predict a system state vector of the dynamical system in time based on a current scenario parameter vector and a control vector
  • said control heuristic generation system comprises: a first computation unit using a model predictive control, MPC, algorithm to provide the control vector at every time during a simulation of said dynamical system using said simulation model f for different scenario parameter vectors and initial system state vectors and adapted to calculate for every simulated combination of scenario parameter vector and initial system state vector a resulting optimal control value using said Model Predictive Control, MPC, algorithm and saving the resulting optimal control value in a memory, a second computational unit adapted to generate machine learned control heuristics approximating the relationship between the corresponding scenario parameter vector and the initial system state vector for the saved resulting optimal control value using a machine learning algorithm, wherein
  • control heuristic generation platform is implemented as a cloud platform.
  • control heuristic generation platform in a further possible embodiment of the control heuristic generation platform according to the second aspect of embodiments of the present invention the machine learned control heuristics comprise approximate rules for controlling the complex dynamical system modelled by said simulation model f.
  • the dynamical system comprises a vehicle controlled online by a controller according to the transferred machine learned control heuristics.
  • FIG. 1 shows schematically a possible exemplary embodiment of a control heuristic generation platform according to an aspect of embodiments of the present invention
  • FIG. 2 shows a further schematic diagram for illustrating a possible exemplary embodiment of a control heuristic generation platform providing machine learned control heuristics used by a controller for controlling a dynamical system;
  • FIG. 3 shows a flow chart of a possible exemplary embodiment of a method for performing an optimized control of a complex dynamical system according to an aspect of embodiments of the present invention
  • FIG. 4A, 4B, 4C, 4D show diagrams for illustrating possible exemplary embodiments for a method for performing an optimized control of a complex system according to an aspect of embodiments of the present invention
  • FIG. 5 shows a diagram of model predictive control results for illustrating a possible use case of the system according to embodiments of the present invention.
  • FIG. 6 shows a possible system's response for the use case of FIG. 5 .
  • a control heuristic generation platform 1 can comprise different components.
  • the control heuristic generation platform 1 as illustrated in FIG. 1 can be used for providing machine learned control heuristics. These control heuristics are used for controlling a dynamical system sys online, which is modelled by a simulation model f and stored in model storage 2 of the control heuristic generation platform 1 .
  • the simulation model f stored in the model storage 2 is adapted to predict a system state vector x of the dynamical system sys in time based on a current scenario parameter vector p and a control vector u.
  • the simulation model f can be an engineering model of the dynamical system sys.
  • the dynamical system can be for instance a vehicle driving on a road.
  • the control heuristic generation platform 1 comprises a first computation unit 3 adapted to use a model predictive control, MPC, algorithm to provide the control vector u at every time during a simulation of the dynamical system using the simulation model f for different scenario parameter vectors p and initial system state vectors x 0 .
  • the first computation unit 3 of the control heuristic generation platform 1 has access to a database 4 which stores scenario parameter vectors p. Further, the computation unit 3 of the control heuristic generation platform 1 has access to a further database 5 where initial system state vectors x 0 of the system sys are stored.
  • the computation unit 3 is adapted to calculate for every simulated combination of scenario parameter vector p and initial system state vector x 0 a resulting optimal control value u* using the MPC algorithm.
  • the resulting optimal control value u* is then saved by the computation unit 3 to a memory 6 which stores the calculated optimal control values.
  • the control heuristic generation platform 1 further comprises a second computation unit 7 adapted to generate machine learned control heuristics approximating a relationship between the corresponding scenario parameter vector p and the initial system state vector x 0 for the resulting optimal control value using a machine learning algorithm MLA.
  • the generated machine learned control heuristics u a can be transferred in a possible embodiment to a controller 9 of the dynamical system sys via an interface of the control heuristic generation platform 1 . As shown in FIG. 1 the generated machine learned control heuristics u a are after completion transferred via means 8 to the controller 9 controlling the dynamical system sys modelled by the simulation model f online.
  • control heuristic generation platform 1 can be implemented as a remote platform, for instance as a cloud platform.
  • the implementation as a cloud platform allows using almost unlimited computation resources in the offline, heuristic-producing phase.
  • FIG. 2 shows a further diagram for illustrating the system according to embodiments of the present invention.
  • the remote platform 1 comprises a storage 2 for storing the engineering model or simulation model f of the dynamical system sys.
  • the platform 1 comprises in the illustrated embodiment a model order reduction unit 2 a for reducing automatically the complexity or order of the used simulation model f.
  • the database 4 stores a group of anticipated use cases or scenario parameter vectors, for instance historic data, previous product generations or product requirements.
  • the computation unit 3 performs simulations using a model predictive control, MPC, to provide a control vector u. Accordingly, the computation unit 3 provides a model predictive control, MPC, for the group of anticipated scenarios.
  • the model predictive control, MPC, algorithm is used to provide a control vector u at every time during a simulation of the dynamical system sys using the simulation model f for different scenario parameter vectors p and initial system state vectors x 0 read from the database 5 .
  • a resulting optimal control value u* is calculated using the MPC algorithm.
  • the resulting optimal control values u* are stored in the memory 6 of the platform 1 .
  • the second computation unit 7 of the platform 1 generates machine learned control heuristics u a approximating the relationship between the corresponding scenario parameter vector p and the initial system state vector x 0 for the saved resulting optimal control values u* using a machine learning algorithm MLA.
  • Control values are learned by means of machine learning on the basis of the stored control behavior and/or control curves for instance by means of neural networks.
  • the generated machine learned control heuristics u a can then be transferred to an internal memory 9 A of the controller 9 as illustrated in FIG. 2 .
  • the control heuristic generation platform 1 performs an offline pre-processing using machine learning based on a simulation model f, a set of parameter scenarios using a model predictive control, MPC, algorithm. Accordingly, the control heuristic generation platform 1 provides a pre-processing in an offline phase, i.e. before operation of the system sys.
  • the system sys can include one or several processes 10 A as shown in FIG. 2 .
  • a process of the system can be controlled using the control heuristics u a based on a predicted load 9 B at a current state of the process as also illustrated in FIG. 2 .
  • Additional use cases or scenarios can be stored in the database 4 of the platform 1 as illustrated in FIG. 2 . These additional use cases or scenarios may be gathered by saving parameter and state values during the online phase. It is also possible that an update of the simulation model f is performed on the basis of additional sensor information such as the change of parameters through degradation.
  • a current load (current value of some scenario parameters) 10 B is supplied to the process 10 A of the dynamical system sys as shown in FIG. 2 .
  • Input parameters of the simulation model can be calibrated leading to better simulation models, e.g. learning how wear impacts motor properties, detecting tires with less pressure than assumed, detecting roof top boxed leading to a different aerodynamics, . . .
  • An example of a dynamical system sys is a vehicle driving along a hilly road
  • the state vector entries of the state vector x can be quantities such as the height, inclination, position, velocity, acceleration of the vehicle as well as the angular velocity of the motor of the vehicle.
  • the output vector entries can comprise the readings of speed and consumption on a dashboard of the vehicle. From these the state vector x can be inferred either directly or indirectly.
  • the control vector entries of the control vector u can for instance indicate how much gas is supplied and how far the brake pedals are pressed down.
  • the parameter vector entries of the parameter vector p can for instance comprise the mass, air drag and rolling resistance profiles as well as the motor torque profile of the vehicle. Further parameter vector entries can describe a height profile of the road such as tabulated values of height with a position or coefficients of some function describing the height curve.
  • the Model Predictive Control is an algorithm for calculating an optimal way of controlling a controllable dynamical system sys.
  • T h is called the prediction horizon length.
  • FIG. 3 shows a flow chart of a possible exemplary embodiment of a method for performing an optimized control of a complex dynamical system sys according to an aspect of embodiments of the present invention.
  • a simulation model f is provided for predicting a system state vector x of the dynamical system in time based on a scenario parameter vector p and a control vector u.
  • a model predictive control, MPC, algorithm is used to provide the control vector u at every time during a simulation of the dynamical system using the simulation model f for different scenario parameter vectors p and initial system state vectors x 0 .
  • a resulting optimal control value u* is calculated by the MPC algorithm and saved to a memory.
  • machine learned control heuristics u a are generated approximating the relationship between corresponding scenario parameter vector p and the initial system state vector x 0 for the saved resulting optimal control value u* using a machine learning algorithm MLA.
  • a machine learning algorithm MLA can use for instance diffusion maps.
  • the machine learning algorithm MLA uses diffusion maps with closed observables for approximating the dynamical system sys.
  • step S 5 the generated machine learned heuristics u a are used to control the complex dynamical system sys modelled by the simulation model f online.
  • FIGS. 4A to 4D show steps of a method according to embodiments of the present invention.
  • high-dimensional MPC data is then parameterized on a lower dimensional manifold by using diffusion maps resulting in diffusion map coordinates ⁇ 0 and ⁇ 1 .
  • a diffusion map is a data analysing tool for parameterizing lower dimensional manifolds embedded in higher dimensions. Mapping according to Diffusion Maps is based upon distances between near neighbours, measured according to some kernel or metric. The idea underlying diffusion maps is that for points sampled in an underlying manifold the distance to the closest ones in the high dimensional space will approximately be the same as the distance in or along the manifold. This can be used for a type of diffusion operator between the points which in the limit of unlimited data can be made to approximate a continuous Laplace-Beltrami operator.
  • eigenvectors or eigenfunctions of this operator provides useful parameterizations of the underlying manifold, it is possible to do the same for the discrete one and so to obtain a parameterization.
  • the eigenvector values can be interpreted as coordinates in the most significant directions, or the directions that provide the most contribution to the distance within the manifold.
  • eigenvectors they also form an orthogonal basis of the space of functions on the data. This can also form the basis for interpolating and extending functions through geometrics harmonics.
  • the diffusion map's coordinates can be extended and interpolated to non-sampled points lying on or near the low-dimensional manifold approximated by the diffusion maps.
  • the output heuristics can be interpolated at previously unsampled points and can then be used for example for validation or control.
  • Another possibility of incorporating the dynamics of the system is by expanding each data point through time-delayed embedding, that is, incorporating a whole time series as one data point. This way, the distance metric compares differences between whole trajectories rather than single points in time.
  • the complex dynamical system comprises a vehicle driving on a road.
  • a system includes the energy optimized acceleration and break of the vehicle regarding a pre-provided height profile (scenario p) based on a current speed (state x).
  • a control heuristic can be trained by a height profile of map routes (such as those provided by Google Maps or other map services and navigation systems) as illustrated in FIG. 5 .
  • a model predictive control (MPC) algorithm can deliver the optimized control strategies or optimized control values for this height profile.
  • FIG. 5 shows the height profile, the velocity V, acceleration A, temperature T and a motor breaking force F as well as an instantaneous cost/time value c, over a distance d.
  • FIG. 6 A system response with optimal control values from MPC to two successive hills made up by gaussians of a standard deviation 1000 height 200 and centres 3000 apart is shown in FIG. 6 .
  • the control heuristic generation platform 1 comprises a database 4 for storing scenario parameter vectors.
  • a scenario parameter vector is a vector of parameters that describe the external factors of the system. These external factors do not change because of the system's evolution. These parameters can comprise process requirements, physical constants, properties of the system or properties of a specific setting.
  • the model predictive control MPC is used to provide the control vector u at every time during a simulation of the dynamical system sys using the simulation model f for different scenario parameter vectors p and initial system state vectors x 0 .
  • a resulting optimal control value u* is calculated and saved in a memory 6 as shown in FIG. 1 .
  • a machine learning algorithm MLA is used to generate machine learned control heuristics u a approximating the relationship between the corresponding scenario parameter vector p and the initial system state vector x 0 for the saved resulting optimal control value u*.
  • a machine learned heuristic is a heuristic (a rule of thumb or approximative rule) generated by a machine learning algorithm MLA. Machine learned heuristics comprise therefore approximate rules for control.
  • the machine learned control heuristics u a refer to an approximative rule on the input/output relation from the input system vector x and scenario parameter vector p to the optimal control vector u generated by the model predictive control MPC algorithm. Consequently, the machine learned control heuristic u a as such is a mathematical function taking the input state vector x and the scenario parameter vector p and outputting the control vector u. Control heuristics provide a computed optimal control value. The resulting optimal control values u* can be saved and then a relationship (heuristic) can be interpolated.
  • linear regression or nearest neighbour interpolation can be used to produce an approximate rule on variables.
  • the system sys itself can be controlled during operation by the controller 9 using only the generated machine learned control heuristics u a . Accordingly, the computational requirements of the controller 9 itself are low. Further, the control is robust in comparison to conventional model predictive controls MPCs. Even complex processes or systems sys can be represented by simple control heuristics u a so that the efficiency of the control method is increased.

Landscapes

  • Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Automation & Control Theory (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Human Computer Interaction (AREA)
  • Mechanical Engineering (AREA)
  • Transportation (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Feedback Control In General (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)

Abstract

A method using machine learned, scenario based control heuristics including: providing a simulation model for predicting a system state vector of the dynamical system in time based on a current scenario parameter vector and a control vector; using a Model Predictive Control, MPC, algorithm to provide the control vector during a simulation of the dynamical system using the simulation model for different scenario parameter vectors and initial system state vectors; calculating a scenario parameter vector and initial system state vector a resulting optimal control value by the MPC algorithm; generating machine learned control heuristics approximating the relationship between the corresponding scenario parameter vector and the initial system state vector for the resulting optimal control value using a machine learning algorithm; and using the generated machine learned control heuristics to control the complex dynamical system modelled by the simulation model.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS
This application claims priority to European application No. EP 17171020 having a filing date of May 15, 2017, the entire contents of which is hereby incorporated by reference.
FIELD OF TECHNOLOGY
The following relates to a system and method for providing optimized control of a complex dynamical system such as a vehicle using machine learned, scenario based control heuristics.
BACKGROUND
Systems such as plants or vehicles are becoming increasingly more complex. As a consequence, also the corresponding controls controlling the system become more complex. This leads to increasing requirements for programming and configuration of corresponding control systems. Moreover, the computational requirements have become more demanding and the necessary computation time of performing a system control does increase.
SUMMARY
An aspect relates to providing a method system for controlling a complex dynamical system which is highly efficient and requires less computational resources for performing the control of the complex dynamical system.
Embodiments of the present invention provide according to a first aspect a method for performing an optimized control of a complex dynamical system using machine learned, scenario based control heuristics. The method comprising the steps of: providing a simulation model for predicting a system state vector of said dynamical system in time based on a current scenario parameter vector and a control vector, using a Model Predictive Control, MPC, algorithm to provide the control vector at every time during a simulation of the dynamical system using said simulation model for different scenario parameter vectors and initial system state vectors, calculating for every simulated combination of scenario parameter vector and initial system state vector a resulting optimal control value by the Model Predictive Control, MPC, algorithm and saving the resulting optimal control value, generating machine learned control heuristics approximating the relationship between the corresponding scenario parameter vector and the initial system state vector for the saved resulting optimal control value using a machine learning algorithm, and using the generated machine learned control heuristics to control the complex dynamical system modelled by said simulation model.
In a possible embodiment of the method according to the first aspect of embodiments of the present invention the machine learning algorithm uses diffusion maps.
In another possible embodiment of the method according to the first aspect of embodiments of the present invention the machine learning algorithm uses diffusion maps with closed observables for approximating the dynamical system.
In another possible embodiment of the method according to the first aspect of embodiments of the present invention the machine learning algorithm uses support vector machines.
In a further possible embodiment of the method according to the first aspect of embodiments of the present invention the generated machine learned control heuristics are transferred to a controller which controls online the dynamical system according to the transferred machine learned control heuristics.
In another possible embodiment of the method according to the first aspect embodiments of the present invention the machine learned control heuristics comprise approximate rules for controlling the complex dynamical system modelled by said simulation model.
Embodiments of the present invention provide according to a further aspect a control heuristic generation platform.
Embodiments of the present invention provide according to the second aspect a control heuristic generation platform for providing machine learned control heuristics used for controlling a dynamical system modelled by a simulation model f stored in a model storage and adapted to predict a system state vector of the dynamical system in time based on a current scenario parameter vector and a control vector, wherein said control heuristic generation system comprises: a first computation unit using a model predictive control, MPC, algorithm to provide the control vector at every time during a simulation of said dynamical system using said simulation model f for different scenario parameter vectors and initial system state vectors and adapted to calculate for every simulated combination of scenario parameter vector and initial system state vector a resulting optimal control value using said Model Predictive Control, MPC, algorithm and saving the resulting optimal control value in a memory, a second computational unit adapted to generate machine learned control heuristics approximating the relationship between the corresponding scenario parameter vector and the initial system state vector for the saved resulting optimal control value using a machine learning algorithm, wherein the generated machine learned control heuristics are transferable to a controller of said dynamical system via an interface of said control heuristic generation platform.
In a possible embodiment of the control heuristic generation platform according to the second aspect of embodiments of the present invention the control heuristic generation platform is implemented as a cloud platform.
In a further possible embodiment of the control heuristic generation platform according to the second aspect of embodiments of the present invention the machine learned control heuristics comprise approximate rules for controlling the complex dynamical system modelled by said simulation model f.
In a further possible embodiment of the control heuristic generation platform according to the second aspect of embodiments of the present invention the dynamical system comprises a vehicle controlled online by a controller according to the transferred machine learned control heuristics.
BRIEF DESCRIPTION
Some of the embodiments will be described in detail, with reference to the following figures, wherein like designations denote like members, wherein:
FIG. 1 shows schematically a possible exemplary embodiment of a control heuristic generation platform according to an aspect of embodiments of the present invention;
FIG. 2 shows a further schematic diagram for illustrating a possible exemplary embodiment of a control heuristic generation platform providing machine learned control heuristics used by a controller for controlling a dynamical system;
FIG. 3 shows a flow chart of a possible exemplary embodiment of a method for performing an optimized control of a complex dynamical system according to an aspect of embodiments of the present invention;
FIG. 4A, 4B, 4C, 4D show diagrams for illustrating possible exemplary embodiments for a method for performing an optimized control of a complex system according to an aspect of embodiments of the present invention;
FIG. 5 shows a diagram of model predictive control results for illustrating a possible use case of the system according to embodiments of the present invention; and
FIG. 6 shows a possible system's response for the use case of FIG. 5.
 DETAILED DESCRIPTION
As can be seen in the exemplary embodiment of FIG. 1 a control heuristic generation platform 1 according to an aspect of embodiments of the present invention can comprise different components. The control heuristic generation platform 1 as illustrated in FIG. 1 can be used for providing machine learned control heuristics. These control heuristics are used for controlling a dynamical system sys online, which is modelled by a simulation model f and stored in model storage 2 of the control heuristic generation platform 1. The simulation model f stored in the model storage 2 is adapted to predict a system state vector x of the dynamical system sys in time based on a current scenario parameter vector p and a control vector u. The simulation model f can be an engineering model of the dynamical system sys. The dynamical system can be for instance a vehicle driving on a road. The control heuristic generation platform 1 comprises a first computation unit 3 adapted to use a model predictive control, MPC, algorithm to provide the control vector u at every time during a simulation of the dynamical system using the simulation model f for different scenario parameter vectors p and initial system state vectors x0. The first computation unit 3 of the control heuristic generation platform 1 has access to a database 4 which stores scenario parameter vectors p. Further, the computation unit 3 of the control heuristic generation platform 1 has access to a further database 5 where initial system state vectors x0 of the system sys are stored. The computation unit 3 is adapted to calculate for every simulated combination of scenario parameter vector p and initial system state vector x0 a resulting optimal control value u* using the MPC algorithm. The resulting optimal control value u* is then saved by the computation unit 3 to a memory 6 which stores the calculated optimal control values.
The control heuristic generation platform 1 further comprises a second computation unit 7 adapted to generate machine learned control heuristics approximating a relationship between the corresponding scenario parameter vector p and the initial system state vector x0 for the resulting optimal control value using a machine learning algorithm MLA. The generated machine learned control heuristics ua can be transferred in a possible embodiment to a controller 9 of the dynamical system sys via an interface of the control heuristic generation platform 1. As shown in FIG. 1 the generated machine learned control heuristics ua are after completion transferred via means 8 to the controller 9 controlling the dynamical system sys modelled by the simulation model f online. They can be transported by different means, such as electronically by cable or by a physical memory such as a USB stick. In a possible implementation the control heuristic generation platform 1 can be implemented as a remote platform, for instance as a cloud platform. The implementation as a cloud platform allows using almost unlimited computation resources in the offline, heuristic-producing phase.
FIG. 2 shows a further diagram for illustrating the system according to embodiments of the present invention. In the illustrated embodiment of FIG. 2 the remote platform 1 comprises a storage 2 for storing the engineering model or simulation model f of the dynamical system sys. The platform 1 comprises in the illustrated embodiment a model order reduction unit 2 a for reducing automatically the complexity or order of the used simulation model f. The database 4 stores a group of anticipated use cases or scenario parameter vectors, for instance historic data, previous product generations or product requirements. The computation unit 3 performs simulations using a model predictive control, MPC, to provide a control vector u. Accordingly, the computation unit 3 provides a model predictive control, MPC, for the group of anticipated scenarios. The model predictive control, MPC, algorithm is used to provide a control vector u at every time during a simulation of the dynamical system sys using the simulation model f for different scenario parameter vectors p and initial system state vectors x0 read from the database 5. On every simulation combination of a scenario parameter vector p and initial system state vector x0 a resulting optimal control value u* is calculated using the MPC algorithm. The resulting optimal control values u* are stored in the memory 6 of the platform 1. The second computation unit 7 of the platform 1 generates machine learned control heuristics ua approximating the relationship between the corresponding scenario parameter vector p and the initial system state vector x0 for the saved resulting optimal control values u* using a machine learning algorithm MLA. Control values are learned by means of machine learning on the basis of the stored control behavior and/or control curves for instance by means of neural networks. The generated machine learned control heuristics ua can then be transferred to an internal memory 9A of the controller 9 as illustrated in FIG. 2. The control heuristic generation platform 1 performs an offline pre-processing using machine learning based on a simulation model f, a set of parameter scenarios using a model predictive control, MPC, algorithm. Accordingly, the control heuristic generation platform 1 provides a pre-processing in an offline phase, i.e. before operation of the system sys.
After the control heuristics ua have been transferred to the internal memory 9A of the controller 9 an online control phase is initiated during operation of the system sys. The system sys can include one or several processes 10A as shown in FIG. 2. A process of the system can be controlled using the control heuristics ua based on a predicted load 9B at a current state of the process as also illustrated in FIG. 2. Additional use cases or scenarios can be stored in the database 4 of the platform 1 as illustrated in FIG. 2. These additional use cases or scenarios may be gathered by saving parameter and state values during the online phase. It is also possible that an update of the simulation model f is performed on the basis of additional sensor information such as the change of parameters through degradation. A current load (current value of some scenario parameters) 10B is supplied to the process 10A of the dynamical system sys as shown in FIG. 2.
There is a feedback of scenarios learned according to the operation to the data base. This feedback improves the control significantly over time.
Based on this data,
A) Input parameters of the simulation model can be calibrated leading to better simulation models, e.g. learning how wear impacts motor properties, detecting tires with less pressure than assumed, detecting roof top boxed leading to a different aerodynamics, . . .
B) If can be determined under which loads the system, e.g. a machine is operated. In the case of a car that can be done by reading the trajectory from a map. In the case of a ship, factory or pump thin can be more complex. E.g. it can be determined for a pump what kind of oil with different viscosities is pumped through a pipeline. E.g. different viscosities can mean different loads.
A dynamical system is a system which has a state vector (x), where the state vector evolves with time t according to some function of the state, f(x)=dx/dt. Additionally, the function can accept some parameters in a vector p (dx/dt=f(x, p)). In a controllable dynamical system an additional control vector (u) represents a part of the system that can be changed directly by a controller 9 that may influence the evolution of the dynamical system (dx/dt=f(x, u, p)). An example of a dynamical system sys is a vehicle driving along a hilly road where the state vector entries of the state vector x can be quantities such as the height, inclination, position, velocity, acceleration of the vehicle as well as the angular velocity of the motor of the vehicle. The output vector entries can comprise the readings of speed and consumption on a dashboard of the vehicle. From these the state vector x can be inferred either directly or indirectly. The control vector entries of the control vector u can for instance indicate how much gas is supplied and how far the brake pedals are pressed down. The parameter vector entries of the parameter vector p can for instance comprise the mass, air drag and rolling resistance profiles as well as the motor torque profile of the vehicle. Further parameter vector entries can describe a height profile of the road such as tabulated values of height with a position or coefficients of some function describing the height curve.
The Model Predictive Control (MPC) is an algorithm for calculating an optimal way of controlling a controllable dynamical system sys. A Model Predictive Control (MPC) takes as an input the evolution function, the estimated state vector x at time t=0 and the parameter vector p, possibly some constraints d(x, u, p), as well as a cost function C(t, x, u, p), that shall be as low as possible, and uses an optimisation algorithm to find an optimal control u* from t=0 to t=Th, that either exactly or approximately gives the lowest value of the cost function C integrated during this time period, which is called the prediction horizon, while fulfilling the constraints if possible. Th is called the prediction horizon length. This calculated optimal control u* is implemented for a time step ΔTc, where 0<ΔTc<Th, after which the optimal control u* is found again for the time from t=ΔTc to t=Th+ΔTc, using the new estimates of the state vector x at time t=ΔTc. This can then be repeated for the next time interval from ΔTc to 2 ΔTc and so on for as long as desired.
FIG. 3 shows a flow chart of a possible exemplary embodiment of a method for performing an optimized control of a complex dynamical system sys according to an aspect of embodiments of the present invention.
In a first step S1 a simulation model f is provided for predicting a system state vector x of the dynamical system in time based on a scenario parameter vector p and a control vector u.
In a further step S2 a model predictive control, MPC, algorithm is used to provide the control vector u at every time during a simulation of the dynamical system using the simulation model f for different scenario parameter vectors p and initial system state vectors x0.
In a further step S3 for every simulated combination of the scenario parameter vector p and initial system state vector x0 a resulting optimal control value u* is calculated by the MPC algorithm and saved to a memory.
In a further step S4 machine learned control heuristics ua are generated approximating the relationship between corresponding scenario parameter vector p and the initial system state vector x0 for the saved resulting optimal control value u* using a machine learning algorithm MLA. A machine learning algorithm MLA can use for instance diffusion maps. In a possible embodiment the machine learning algorithm MLA uses diffusion maps with closed observables for approximating the dynamical system sys.
In a further step S5 the generated machine learned heuristics ua are used to control the complex dynamical system sys modelled by the simulation model f online.
The controllable dynamical system sys can be represented by a simulation model f for the dynamics of the system as follows:
{dot over (x)}=f(t,x,u,p)
wherein t is a time, x is a state variable vector, u is a control variable vector and p is a parameter vector.
FIGS. 4A to 4D show steps of a method according to embodiments of the present invention.
FIG. 4A shows an MPC optimization for one-dimensional states x (t) and controls u (t). Based on known data for x and u (solid lines) and the dynamics x=f(x, u) of the system, a cost function c (not shown) is minimized for a prediction horizon t∈[t0, t0+Thor]. The resulting optimal control values for u and corresponding predicted states x (dashed lines) can then be used for the next ΔTc until t1, at which point the procedure is repeated for the interval t∈[t0, t0+Thor].
As illustrated in FIG. 4B high-dimensional MPC data is then parameterized on a lower dimensional manifold by using diffusion maps resulting in diffusion map coordinates ψ0 and ψ1. A diffusion map is a data analysing tool for parameterizing lower dimensional manifolds embedded in higher dimensions. Mapping according to Diffusion Maps is based upon distances between near neighbours, measured according to some kernel or metric. The idea underlying diffusion maps is that for points sampled in an underlying manifold the distance to the closest ones in the high dimensional space will approximately be the same as the distance in or along the manifold. This can be used for a type of diffusion operator between the points which in the limit of unlimited data can be made to approximate a continuous Laplace-Beltrami operator. As the eigenvectors or eigenfunctions of this operator provides useful parameterizations of the underlying manifold, it is possible to do the same for the discrete one and so to obtain a parameterization. The eigenvector values can be interpreted as coordinates in the most significant directions, or the directions that provide the most contribution to the distance within the manifold. As eigenvectors they also form an orthogonal basis of the space of functions on the data. This can also form the basis for interpolating and extending functions through geometrics harmonics.
As illustrated in FIG. 4C the diffusion map's coordinates can be extended and interpolated to non-sampled points lying on or near the low-dimensional manifold approximated by the diffusion maps.
Further, as illustrated in FIG. 4D the output heuristics can be interpolated at previously unsampled points and can then be used for example for validation or control.
Another possibility of incorporating the dynamics of the system is by expanding each data point through time-delayed embedding, that is, incorporating a whole time series as one data point. This way, the distance metric compares differences between whole trajectories rather than single points in time.
In a possible embodiment the complex dynamical system comprises a vehicle driving on a road. In an application example such a system includes the energy optimized acceleration and break of the vehicle regarding a pre-provided height profile (scenario p) based on a current speed (state x).
For example, a control heuristic can be trained by a height profile of map routes (such as those provided by Google Maps or other map services and navigation systems) as illustrated in FIG. 5. A model predictive control (MPC) algorithm can deliver the optimized control strategies or optimized control values for this height profile. FIG. 5 shows the height profile, the velocity V, acceleration A, temperature T and a motor breaking force F as well as an instantaneous cost/time value c, over a distance d.
A system response with optimal control values from MPC to two successive hills made up by gaussians of a standard deviation 1000 height 200 and centres 3000 apart is shown in FIG. 6.
The control heuristic generation platform 1 comprises a database 4 for storing scenario parameter vectors. A scenario parameter vector is a vector of parameters that describe the external factors of the system. These external factors do not change because of the system's evolution. These parameters can comprise process requirements, physical constants, properties of the system or properties of a specific setting.
Further the platform 1 has access to initial system state vectors of the system. A system state vector x is a vector of variables that describe a state of the dynamical system sys and which influence the future of the dynamical system sys and that do also evolve with time. Evolution is described by a mathematical model f that depends on the system state vector x, the vector of scenario parameters p and the vector of control variables u according to dx/dt=f(x, u, p). The model predictive control MPC is used to provide the control vector u at every time during a simulation of the dynamical system sys using the simulation model f for different scenario parameter vectors p and initial system state vectors x0. For every simulated combination of a scenario parameter vector p and an initial system state vector x0 a resulting optimal control value u* is calculated and saved in a memory 6 as shown in FIG. 1. A machine learning algorithm MLA is used to generate machine learned control heuristics ua approximating the relationship between the corresponding scenario parameter vector p and the initial system state vector x0 for the saved resulting optimal control value u*. A machine learned heuristic is a heuristic (a rule of thumb or approximative rule) generated by a machine learning algorithm MLA. Machine learned heuristics comprise therefore approximate rules for control. The machine learned control heuristics ua refer to an approximative rule on the input/output relation from the input system vector x and scenario parameter vector p to the optimal control vector u generated by the model predictive control MPC algorithm. Consequently, the machine learned control heuristic ua as such is a mathematical function taking the input state vector x and the scenario parameter vector p and outputting the control vector u. Control heuristics provide a computed optimal control value. The resulting optimal control values u* can be saved and then a relationship (heuristic) can be interpolated.
In a further possible embodiment, also linear regression or nearest neighbour interpolation can be used to produce an approximate rule on variables.
The system sys itself can be controlled during operation by the controller 9 using only the generated machine learned control heuristics ua. Accordingly, the computational requirements of the controller 9 itself are low. Further, the control is robust in comparison to conventional model predictive controls MPCs. Even complex processes or systems sys can be represented by simple control heuristics ua so that the efficiency of the control method is increased.
Although the present invention has been disclosed in the form of preferred embodiments and variations thereon, it will be understood that numerous additional modifications and variations could be made thereto without departing from the scope of the invention.
For the sake of clarity, it is to be understood that the use of “a” or “an” throughout this application does not exclude a plurality, and “comprising” does not exclude other steps or elements.

Claims (11)

The invention claimed is:
1. A method for performing an optimized control of a complex dynamical system using machine learned, scenario based control heuristics, the method comprising the steps of:
a) providing a simulation model for predicting a system state vector of said dynamical system in time based on a current scenario parameter vector and a control vector;
b) using a Model Predictive Control, MPC, algorithm to provide the control vector at every time during a simulation of said dynamical system using said simulation model for different scenario parameter vectors and initial system state vectors;
c) calculating for every simulated combination of a scenario parameter vector and initial system state vector a resulting optimal control value by the MPC algorithm and saving the resulting optimal control value;
d) generating machine learned control heuristics approximating the relationship between the corresponding scenario parameter vectors and the initial system state vectors for the saved resulting optimal control values using a machine learning algorithm; and
e) using the generated machine learned control heuristics to control the complex dynamical system modelled by said simulation model.
2. The method according to claim 1 wherein the machine learning algorithm uses diffusion maps.
3. The method according to claim 1 wherein the machine learning algorithm uses diffusion maps with closed observables for approximating said dynamical system.
4. The method according to claim 1 wherein the machine learning algorithm uses support vector machines.
5. The method according to claim 1 wherein the generated machine learned control heuristics are transferred to a controller which controls online the dynamical system according to the transferred machine learned control heuristics.
6. The method according to claim 1 wherein the machine learned control heuristics comprise approximate rules for controlling the complex dynamical system modelled by said simulation model.
7. A control heuristic generation platform for providing machine learned control heuristics used for controlling a dynamical system modelled by a simulation model stored in a model storage and adapted to predict a system state vector of said dynamical system in time based on a current scenario parameter vector and a control vector, wherein said control heuristic generation system comprises:
a first computation unit using a model predictive control, MPC, algorithm to provide the control vector at every time during a simulation of said dynamical system using said simulation model for different scenario parameter vectors and initial system state vectors and adapted to calculate for every simulated combination of scenario parameter vector and initial system state vector a resulting optimal control value by the MPC algorithm and saving the resulting optimal control value to a memory;
a second computational unit adapted to
generate machine learned control heuristics approximating the relationship between the corresponding scenario parameter vector and the initial system state vector for the saved resulting optimal control value using a machine learning algorithm, wherein the generated machine learned control heuristics are transferable to a controller of said dynamical system via an interface of said control heuristic generation platform.
8. The control heuristic generation platform according to claim 7, wherein initial system states and scenario parameter vectors observed and saved during online operation are additionally supplied to the first computational unit to construct updated machine learned control heuristics that are transferable to a controller.
9. The control heuristic generation platform according to claim 7, wherein the control heuristic generation platform is implemented as a cloud platform.
10. The control heuristic generation platform according to claim 7 wherein the machine learned control heuristics comprise approximate rules for controlling the complex dynamical system modelled by said simulation model.
11. The control heuristic generation platform according to claim 7, wherein the dynamical system comprises a vehicle controlled online by said controller according to the transferred machine learned control heuristics.
US15/963,240 2017-05-15 2018-04-26 Method and system for providing an optimized control of a complex dynamical system Expired - Fee Related US10953891B2 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
EP17171020.5A EP3404497B1 (en) 2017-05-15 2017-05-15 A method and system for providing an optimized control of a complex dynamical system
EP17171020 2017-05-15

Publications (2)

Publication Number Publication Date
US20190031204A1 US20190031204A1 (en) 2019-01-31
US10953891B2 true US10953891B2 (en) 2021-03-23

Family

ID=58709347

Family Applications (1)

Application Number Title Priority Date Filing Date
US15/963,240 Expired - Fee Related US10953891B2 (en) 2017-05-15 2018-04-26 Method and system for providing an optimized control of a complex dynamical system

Country Status (4)

Country Link
US (1) US10953891B2 (en)
EP (1) EP3404497B1 (en)
JP (1) JP6564905B2 (en)
CN (1) CN108873692B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110032075B (en) * 2019-05-28 2022-04-05 哈尔滨工程大学 A Design Method of Saturation Compensation Control System for Dynamic Positioning Vessel
JP7441115B2 (en) * 2020-05-26 2024-02-29 Ｔｏｙｏ Ｔｉｒｅ株式会社 Tire design support methods, systems and programs
CN116113893B (en) * 2020-07-29 2025-06-17 西门子工业软件有限责任公司 Controlling technical systems with the aid of data-based control models
CN112116140B (en) * 2020-09-10 2022-09-20 同济大学 Building energy consumption prediction method based on twin model
CN113627087B (en) * 2021-08-23 2023-11-03 中机中联工程有限公司 Production system parameter self-adaptive optimization method based on logistics simulation and machine learning
CN115685747B (en) * 2022-09-30 2024-11-22 西南交通大学 A model predictive control method based on residual neural network optimization
US12337870B1 (en) * 2022-09-30 2025-06-24 Zoox, Inc. Heuristic and machine learning agents in driving simulations
CN115447615B (en) * 2022-10-18 2025-09-09 上汽大众汽车有限公司 Track optimization method based on vehicle kinematic model predictive control

Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6445963B1 (en) 1999-10-04 2002-09-03 Fisher Rosemount Systems, Inc. Integrated advanced control blocks in process control systems
US20020184166A1 (en) * 2001-06-04 2002-12-05 Xerox Corporation Method and system for algorithm synthesis in problem solving
US20040249483A1 (en) 2003-06-05 2004-12-09 Wojsznis Wilhelm K. Multiple-input/multiple-output control blocks with non-linear predictive capabilities
US20050267608A1 (en) * 2004-05-27 2005-12-01 Nissan Motor Co., Ltd. Model predictive control apparatus
WO2006026060A1 (en) 2004-08-27 2006-03-09 Alstom Technology Ltd Cascaded control of an average value of a process parameter to a desired value
CN101055468A (en) 2006-04-13 2007-10-17 费舍-柔斯芒特系统股份有限公司 Robust process model identification in model-based control techniques
US20120150324A1 (en) 2010-12-08 2012-06-14 Matthew Brand Method for Solving Control Problems
US20150131848A1 (en) * 2013-11-08 2015-05-14 Analog Devices Technology Support vector machine based object detection system and associated method
CN104967149A (en) 2015-06-29 2015-10-07 山东电力研究院 Micro power grid wind and solar energy storage model prediction control method
US20160147203A1 (en) 2014-11-25 2016-05-26 Mitsubishi Electric Research Laboratories, Inc. Model Predictive Control with Uncertainties
US20160161950A1 (en) 2013-03-19 2016-06-09 Massive Analytic Ltd. Apparatus for controlling a land vehicle which is self-driving or partially self-driving
US20160160787A1 (en) 2014-12-04 2016-06-09 Daimler Ag Controller for controlling an internal combustion engine of a vehicle, in particular a commercial vehicle
WO2017006372A1 (en) 2015-07-08 2017-01-12 Nec Corporation A Method of Optimal Scheduling and Real-time Control for an xManagement System
US20170017212A1 (en) 2014-04-23 2017-01-19 The Florida State University Research Foundation, Adaptive nonlinear model predictive control using a neural network and input sampling

Patent Citations (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6445963B1 (en) 1999-10-04 2002-09-03 Fisher Rosemount Systems, Inc. Integrated advanced control blocks in process control systems
US20020184166A1 (en) * 2001-06-04 2002-12-05 Xerox Corporation Method and system for algorithm synthesis in problem solving
US20040249483A1 (en) 2003-06-05 2004-12-09 Wojsznis Wilhelm K. Multiple-input/multiple-output control blocks with non-linear predictive capabilities
CN1598720A (en) 2003-06-05 2005-03-23 费舍-柔斯芒特系统股份有限公司 Multiple-input/multiple-output control blocks with non-linear predictive capabilities
US20050267608A1 (en) * 2004-05-27 2005-12-01 Nissan Motor Co., Ltd. Model predictive control apparatus
WO2006026060A1 (en) 2004-08-27 2006-03-09 Alstom Technology Ltd Cascaded control of an average value of a process parameter to a desired value
US20060058899A1 (en) 2004-08-27 2006-03-16 Alstom Technology Ltd. Cascaded control of an average value of a process parameter to a desired value
CN105259763A (en) 2006-04-13 2016-01-20 费舍-柔斯芒特系统股份有限公司 Model generation system for generating process model
CN101055468A (en) 2006-04-13 2007-10-17 费舍-柔斯芒特系统股份有限公司 Robust process model identification in model-based control techniques
US20070244575A1 (en) 2006-04-13 2007-10-18 Fisher-Rosemount Systems, Inc. Robust process model identification in model based control techniques
US20120150324A1 (en) 2010-12-08 2012-06-14 Matthew Brand Method for Solving Control Problems
US20160161950A1 (en) 2013-03-19 2016-06-09 Massive Analytic Ltd. Apparatus for controlling a land vehicle which is self-driving or partially self-driving
JP2016520464A (en) 2013-03-19 2016-07-14 マッシブ・アナリティック・リミテッド Device for controlling self-driving or partially self-driving land vehicles
US20150131848A1 (en) * 2013-11-08 2015-05-14 Analog Devices Technology Support vector machine based object detection system and associated method
US20170017212A1 (en) 2014-04-23 2017-01-19 The Florida State University Research Foundation, Adaptive nonlinear model predictive control using a neural network and input sampling
US20160147203A1 (en) 2014-11-25 2016-05-26 Mitsubishi Electric Research Laboratories, Inc. Model Predictive Control with Uncertainties
JP2016100009A (en) 2014-11-25 2016-05-30 三菱電機株式会社 Method for controlling operation of machine and control system for iteratively controlling operation of machine
US20160160787A1 (en) 2014-12-04 2016-06-09 Daimler Ag Controller for controlling an internal combustion engine of a vehicle, in particular a commercial vehicle
CN104967149A (en) 2015-06-29 2015-10-07 山东电力研究院 Micro power grid wind and solar energy storage model prediction control method
WO2017006372A1 (en) 2015-07-08 2017-01-12 Nec Corporation A Method of Optimal Scheduling and Real-time Control for an xManagement System

Non-Patent Citations (10)

* Cited by examiner, † Cited by third party
Title
Afram, Abdul et al: "Artificial neural network (ANN) based model predictive control (MPC) and optimization of HVAC systems: A state of the art review and case study of a residential HVAC system"; Energy and Buildings; vol. 141; Feb. 11, 2017; pp. 9X 113; URL: http://dx.doi.org/10.1016/j.enbuild.2017.02.012; 2017.
English Translation of Japanese Notice of Allowance for Application No. 2018-092884, dated Jul. 2, 2019.
Extended European Search Report for Application No. 17171020.5, dated Nov. 20, 2017.
Gautam Ajay et al: "Stabilizing model predictive control using parameter-dependent dynamic policy for nonlinear systems modeled with neural networks", Journal of Process Control, Oxford, GB, vol. 36, pp. 11-21, XP029309432, ISSN: 0959-1524, DOI: 10.1016/J.JPROCONT.2015.09.003; p. 1-p. 9; 2015.
GAUTAM AJAY; SOH YENG CHAI: "Stabilizing model predictive control using parameter-dependent dynamic policy for nonlinear systems modeled with neural networks", JOURNAL OF PROCESS CONTROL, OXFORD, GB, vol. 36, 5 November 2015 (2015-11-05), GB, pages 11 - 21, XP029309432, ISSN: 0959-1524, DOI: 10.1016/j.jprocont.2015.09.003
Ramirez; "Diffusion maps for exploring electro-optical synthetic vehicle image data"; Aerospace and Electronics Conference (NAECON) 2012 IEEE National; pp. 126-133; 2012 (Year: 2012). *
Sompura Jay N et al: "Experimental study: Neural network based model predictive control of a distributed parameter system", 2016 12th IEEE International Conference on Control and Automation (ICCA), IEEE, pp. 529-534, XP032919061, DOI: 10.1109/ICCA.2016.7505331 Technical Fields [retrieved on Jul. 5, 2016]; 2016.
SOMPURA JAY N.; MANKAD JAIVIK K.; PADHIYAR NITIN U.: "Experimental study: Neural network based model predictive control of a distributed parameter system", 2016 12TH IEEE INTERNATIONAL CONFERENCE ON CONTROL AND AUTOMATION (ICCA), IEEE, 1 June 2016 (2016-06-01), pages 529 - 534, XP032919061, DOI: 10.1109/ICCA.2016.7505331
Xiangjun, Duan: "A study for a predictive control method based on neural networks"; Master's Essay; Mar. 15, 2006; pp. 1-68; 2006.
Yi, He: "Nonlinear system recogonization model and control based on the support vector machine"; The Identification, Modeling and Control of Nonlinear System Based on SVM; Apr. 15, 2009; pp. 1-9; 2009.

Also Published As

Publication number Publication date
US20190031204A1 (en) 2019-01-31
EP3404497A1 (en) 2018-11-21
CN108873692B (en) 2021-07-13
EP3404497B1 (en) 2021-11-10
JP6564905B2 (en) 2019-08-21
CN108873692A (en) 2018-11-23
JP2018195307A (en) 2018-12-06

Similar Documents

Publication Publication Date Title
US10953891B2 (en) Method and system for providing an optimized control of a complex dynamical system
CN113874865B (en) Method and device for determining model parameters of a regulation strategy for a technical system
JP6673293B2 (en) Vehicle system
US11254315B2 (en) Friction adaptive vehicle control
CN113415288B (en) Sectional type longitudinal vehicle speed planning method, device, equipment and storage medium
JP7183446B2 (en) Apparatus and method for controlling machine operation, and storage medium
JP2019066135A (en) Air-conditioning control system
CN111670415A (en) Predictive controller, vehicle and method for control system
CN109415089A (en) Controller and method for controlling a vehicle and non-transitory computer readable memory
WO2018104850A1 (en) Model predictive based control for automobiles
JP2020035182A (en) Control device and control method
TW201322616A (en) Motor control device
WO2018146802A1 (en) Control customization system, control customization method, and control customization program
JPWO2018173933A1 (en) Information processing apparatus, traveling data processing method, vehicle, and program
CN113805483A (en) Robot control method, device and computer equipment based on model prediction
CN117908534A (en) Speed curve optimization method, device, vehicle and storage medium
KR102595615B1 (en) Method for determining safety-critical output values by way of a data analysis device for a technical entity
CN117321511A (en) Robust adaptive dynamic pattern decomposition for modeling, prediction and control of high-dimensional physical systems
CN113939775A (en) Method and device for determining a regulation strategy for a technical system
CN117087669B (en) Vehicle speed tracking control method and device, vehicle and storage medium
CN116501061B (en) A model predictive control path tracking method based on hierarchical optimization
CN116176621A (en) Vehicle dynamic programming control method, device, computer equipment and storage medium
Tran et al. A combination of analytical and model predictive optimal methods for adaptive cruise control problem
CN117962929B (en) Vehicle track tracking control method and device, electronic equipment and storage medium
CN119239643B (en) Vehicle control method and device, vehicle and storage medium

Legal Events

Date Code Title Description
FEPP Fee payment procedure

Free format text: ENTITY STATUS SET TO UNDISCOUNTED (ORIGINAL EVENT CODE: BIG.); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY

AS Assignment

Owner name: SIEMENS AKTIENGESELLSCHAFT, GERMANY

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:HARTMANN, DIRK;OBST, BIRGIT;WANNERBERG, ERIK OLOF JOHANNES;SIGNING DATES FROM 20180516 TO 20180519;REEL/FRAME:046096/0919

STPP Information on status: patent application and granting procedure in general

Free format text: DOCKETED NEW CASE - READY FOR EXAMINATION

STCF Information on status: patent grant

Free format text: PATENTED CASE

FEPP Fee payment procedure

Free format text: MAINTENANCE FEE REMINDER MAILED (ORIGINAL EVENT CODE: REM.); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY

LAPS Lapse for failure to pay maintenance fees

Free format text: PATENT EXPIRED FOR FAILURE TO PAY MAINTENANCE FEES (ORIGINAL EVENT CODE: EXP.); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY

STCH Information on status: patent discontinuation

Free format text: PATENT EXPIRED DUE TO NONPAYMENT OF MAINTENANCE FEES UNDER 37 CFR 1.362

FP Lapsed due to failure to pay maintenance fee

Effective date: 20250323